FEA-1554: Close session data collection gaps across all 5 harness parsers

[thadeusb]

Feature

FEA-1554 — Session Data Collection Gaps for Desktop Session Screens

Parent: PRD-444 (Session Data Collection for Desktop Agent/Session Screens)

Summary

• Extend NormalizedSession + NormalizedToolUse types with 11 new fields for messages, token series, diff stats, slash commands, artifacts, tool output, MCP server/method, skill names
• Update all 5 harness parsers (Claude, Codex, Cursor, Copilot, OpenCode) to extract previously-discarded data: message text, tool output, per-turn tokens, diff stats, artifact references
• Wire cost computation via token-cost.ts into dashboard analytics
• Add event-role.ts (event type → human/agent/system) and session-timing.ts (active/waiting derivation)
• Extend cloud sync STRIPPED_LEAF_KEYS with text, output, reasoning for mandatory content stripping

Architecture

flowchart TD
    Sources[5 Harness Sources] --> Parsers[5 Updated Parsers]
    Parsers --> NS[NormalizedSession + NormalizedToolUse<br/>types.ts — 11 new fields]
    PU[parser-utils.ts<br/>7 new helpers] -.-> Parsers
    NS --> IS[import-session.ts<br/>UserMessage/AssistantMessage events<br/>enriched PostToolUse data]
    IS --> DB[(SQLite: metadata JSON + events.data)]
    DB --> SYNC[STRIPPED_LEAF_KEYS<br/>+text +output +reasoning]
    DB --> COST[dashboard.ts cost wiring]
    DB --> ROLE[event-role.ts + session-timing.ts]

Loading

Plan

PLN-816 — Plan challenge: Opus PASS, Sonnet PASS, Codex PASS

Key Decisions

• Messages stored as individual UserMessage/AssistantMessage events (not metadata JSON) — too large for blob, need idempotency
• Token series stored in metadata (small numeric data, no cloud-strip needed)
• CR-11 role mapping as function not DB column — avoids schema migration
• Cursor diff stats absent at source (verified, recorded [absent])
• Copilot diff stats and reasoning text absent at source (recorded [absent])

Decision Table

Verified at .closedloop-ai/decision-tables/fea-1554.md — 63 delta checklist items covered.

Review Summary

4 parallel review lanes (correctness, security, conventions, codex): all PASS, 0 blockers, 0 findings requiring fixes.

Downstream Updates

None — FEA-1554 is the only child of PRD-444. Sibling features (FEA-1503, FEA-1497) are DONE.

Feature Flags

None — collection-layer improvement, not a net-new user capability. Existing FEA-1407 sandbox gate controls capture.

Database/Migration Safety

No schema/migration changes. All new data stored in existing metadata JSON and events.data JSON columns.

Test Plan

• 49 parser-utils helper tests (truncateText, computeLineDelta, extractRepoFromCwd, extractPrReferences, extractIssueReferences, collectArtifacts, isSyntheticModelKey)
• 14 event-role mapping tests
• 7 session-timing derivation tests
• 4 cloud sync strip tests (text, output, reasoning, round-trip)
• Full suite: 2,263 tests, 0 failures
• Typecheck clean

[github-actions]

Compatibility Smoke Test Results ⚠️

Status: skipped
Electron SHA: 1740b9c5eb92cb3922dc737a72eca49d0bde7e20
Symphony Alpha SHA (last-known-good): 24741ed9be45cc761195d7a7b6613bd30cdfce84
Note: Skipped because the stage GitHub App credentials are not configured for this workflow run.

View Actions run

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 72d023000c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[thadeusb]

Solid extraction work across all 5 parsers and the cloud strip additions are correct. Three things inline, none blocking.